The complaint arrives from a fleet operator in a cold region: a batch of dashcams dead on the coldest mornings of the year. No error screen, no corrupted footage, no low-battery warning. The unit boots normally once the cabin has warmed up, or it wakes mid-drive as if nothing happened. Bench units run all winter without a failure, the supplier finds nothing wrong with returns, and the RMA rate climbs every November. This is not an assembly defect and not a firmware bug. It is a component selection problem, and the component is the quartz crystal that clocks the main SoC.
This build uses the E1SB24E000007E, a 24.000000 MHz seam-sealed SMD crystal unit in a 2016-size package from Hosonic (p. 1). Its electrical parameters are the ones a cold-climate design should be shopping for: fundamental mode, frequency tolerance ±10 ppm at 25 °C, stability ±10 ppm over the operating temperature, an operating range of -30 °C to +85 °C, a 10 pF load capacitance, and a maximum equivalent series resistance of 100 Ω (p. 3). Drive level is 50 µW typical and 100 µW maximum, aging is ±2 ppm in the first year, and the hermetic ceramic package is rated for storage from -55 °C to +125 °C (p. 3).
The verdict up front. For a dashcam mounted inside the cabin where an overnight cold soak stays above -30 °C, this crystal is a straightforward, correct choice: tight tolerance, a range that covers most markets, and a seam-sealed package that shrugs off humidity. It stops being the right answer the moment the mount is a windshield bracket in a market where parked cars drop below -30 °C, because below that floor nothing in this datasheet is guaranteed, and the fix is a different part number, not more board glue.
What the failure looks like in the field
The signature is silence, not distortion. When the oscillator fails to start, the SoC never clocks up, so there is no video artifact, no timestamp glitch and no log entry to inspect. The camera is simply off. That absence of evidence is why these cases drift around support queues for a season before anyone points at timing.
Three details separate it from the usual cold-weather suspects. First, the pattern tracks overnight low temperature, not usage: units parked outside fail, units in garages do not, and the same unit recovers after an hour of cabin heat with no residual fault. Second, the power path is healthy. A weak lithium backup cell produces a different signature, a camera that boots but loses time and settings. Here the time is intact, which means the RTC branch ran and the main clock did not. Third, it reproduces in a chamber and not on a bench, because a bench never sees the temperature the windshield mount sees.
The mount matters more than the climate zone. A windshield bracket soaks to the outside air temperature overnight, below the cabin air, and a unit on the glass starts its morning from the coldest point in the vehicle. The same board on the headliner often survives the same night. Placement changes the effective specification of the crystal without changing a line on the schematic.
Why it happens at the silicon level
The crystal itself is quartz, but the failure mechanism lives at the boundary between the quartz and the oscillator that drives it, so silicon level is the right place to look. The specification guarantees the motional parameters only within the operating range of -30 °C to +85 °C (p. 3). Below the floor, the equivalent series resistance rises steeply as the quartz vibrates less efficiently, and the datasheet makes no promise about how far it climbs.
An oscillator starts only if its negative resistance exceeds the crystal resistance by a comfortable margin. As ESR climbs while temperature falls, that margin erodes from both sides: the crystal demands more gain exactly when the SoC's oscillator transconductance is weakest, because the start-up happens after a cold soak when the die is as cold as the crystal. Below some temperature the loop gain falls short, the oscillation never builds, and the supply current stays flat. The SoC sees a quiet input and the boot never begins.
Frequency is the slower killer. The stability specification of ±10 ppm over the operating range is referred to 25 °C (p. 3), so even when a marginal unit does start, its frequency near the range floor is unguaranteed. A PLL locking to a reference drifting outside its capture window produces intermittent lock, dropped frames or a camera that boots into a watchdog loop. That second signature looks like flaky firmware and comes from the same unguaranteed region of the curve.
The package construction matters here too. The 2016 unit is seam sealed: an alumina ceramic base with a Kovar lid (p. 5), hermetic rather than resin sealed. Resin-sealed crystals breathe with humidity, and moisture that freezes inside the cavity shifts the motional parameters unpredictably. Hermetic sealing removes that variable entirely, worth having in the markets where this failure lives.
The numbers that fix it
| Parameter | Datasheet value | Source |
|---|---|---|
| Nominal frequency | 24.000000 MHz, fundamental mode | p. 1, p. 3 |
| Frequency tolerance | ±10 ppm at 25 °C ± 3 °C | p. 3 |
| Stability over operating temperature | ±10 ppm referred to 25 °C | p. 3 |
| Operating temperature range | -30 °C to +85 °C | p. 3 |
| Storage temperature | -55 °C to +125 °C | p. 3 |
| Load capacitance | 10 pF | p. 3 |
| Equivalent series resistance | 100 Ω maximum | p. 3 |
| Drive level | 50 µW typical, 100 µW maximum | p. 3 |
| Shunt capacitance | 2 pF maximum | p. 3 |
| Insulation resistance | 500 MΩ minimum at 100 V DC | p. 3 |
| Aging | ±2 ppm in the first year | p. 3 |
| Package | 2016 seam-sealed SMD | p. 1 |
The first number to negotiate is the range floor. If the market's overnight lows push the mount below -30 °C, no parameter in this table helps, because the table simply ends there. The fix is a range specified lower, which in the Hosonic catalogue is a part number decision, not a tolerance upgrade. Match the floor to the coldest overnight soak the mount can see, with margin for the windshield effect.
The second pair is ESR and drive level. The 100 Ω maximum is the resistance the oscillator must overcome with margin, and the 50 µW typical to 100 µW maximum drive window is the power budget the resonator tolerates. Both enter the negative resistance calculation, and both are guaranteed only inside the range. Check the SoC's oscillator margin against this resistance limit, not against a number remembered from the previous vendor's crystal.
The third is the frequency budget. Tolerance ±10 ppm, stability ±10 ppm over temperature, and aging ±2 ppm in the first year stack into a total error the SoC's PLL capture range must absorb (p. 3). The remaining lever is the load: this unit wants 10 pF (p. 3), so the two load capacitors are chosen so the crystal sees that value once board strays are added, with the 2 pF maximum shunt capacitance in the calculation (p. 3). A mismatched load pulls the frequency and turns a passing unit marginal without any temperature involved.
Around the family, the same package serves several grades, and each sibling's own datasheet decides which fits. E1SB24E000012E is the nearest twin: same 24 MHz, same ±10 ppm tolerance, same -30 °C to +85 °C range, but ESR at 50 Ω maximum and a 9 pF load, a stiffer resistance spec that buys oscillator margin and a different capacitor trim. E1SB24E000011E is also 50 Ω maximum with an 8 pF load, but its range ends at +80 °C, a reasonable trade for a unit that never sits in windshield sun. E1SB24E00000SE is the indoor grade: ±15 ppm tolerance and stability, a -20 °C to +75 °C range, shunt capacitance up to 3 pF, the part that fails this article's scenario. At the low end, E1SB16E00000FE and E1SB16E000019E share the same seam-sealed 2016 construction; their single-page family specification covers a broad frequency range and quotes a resistance ceiling that loosens as the frequency drops, far more generous than this part's. E1SB20E000000A sits one band up under the same family sheet. E1SB24E000027E is a further ordering code on the same package family, useful when a second sourcing code is wanted.
Verification you can run before the next build
- Reproduce before you fix. Walk a returned unit in a chamber from room temperature to just below the specified floor, dwelling at each step. Log supply current: a crystal that stops starting shows the flat, silent trace long before the unit fails visibly.
- Watch the oscillator pin, not the UART. A scope on the SoC's crystal input during the cold walk shows the start-up transient shrinking as temperature falls, until the ring dies. That point is the real floor of the design, whatever the paper says.
- Compare mounts. A heated stand-in for the cabin versus an unheated stand-in for the windshield separates the placement contribution from the component contribution in one afternoon.
- Measure, do not assume, the load. A capacitance meter across the pads with the crystal removed tells you what the crystal actually sees, including layout strays. Trim the two capacitors until the total lands on the target for the part you ship, because the family trims differ.
- Check the frequency at room temperature with a counter against the ±10 ppm tolerance (p. 3), and recheck after reflow, because the profile allows a peak of 260 °C and up to three passes (p. 6), each stressing the mount.
- Screen incoming stock storage as well as parts. The packing rule keeps storage between -4 °C and 40 °C at 70 percent relative humidity with a 24-month shelf life (p. 3); reels that sat through a humid summer in an unconditioned warehouse deserve a solderability check.
Frequently asked questions
Does swapping in this crystal require board changes?
Only the usual ones for a crystal change. The footprint is the common 2016 pad pattern, but the load capacitance is specified at 10 pF (p. 3), so the two load capacitors must be recalculated for the new shunt capacitance of 2 pF maximum and the board's stray capacitance. Siblings with 8 pF or 9 pF loads need different capacitor values, so a drop-in across the family is a trim change, not a formality.
Which other components cause the same cold symptom?
A weak lithium backup cell produces a cousin of this failure, but with time and settings lost while the camera still boots. Electrolytic capacitors and displays also misbehave in cold soak. The differential is the silence: a crystal start-up failure produces a unit that never clocks, so nothing on the main rail draws its usual current. The supply current signature at low temperature separates the suspects quickly.
How do I verify quickly without a thermal chamber?
A domestic freezer reaches well below the specified floor. Put the unit inside with a battery pack, trigger the boot from a remote command when it comes out, and compare against a unit kept at room temperature. It is crude and will not find the exact margin, but it reliably reproduces a failure that lives below the range floor, enough to confirm the diagnosis before ordering a range-appropriate part.
What does the upgrade cost against the current build?
Within the same package and frequency, price differences between temperature ranges and resistance grades are modest, and the reel size is the same 3,000 pieces per reel (p. 7). The real costs are the capacitor trim if the load changes and one validation pass. Compared with a season of RMAs, the point is that the range floor must match the market.
Sourcing note
The E1SB24E000007E is a Hosonic 2016 seam-sealed crystal unit, 24.000000 MHz fundamental (p. 1), shipped 3,000 pieces per reel, 60,000 per carton (p. 7, p. 8). The reliability program includes thermal shock between -55 °C and +125 °C, vibration from 10 Hz to 2000 Hz, and drop testing from 150 cm (p. 9). For programs that run several years, ask us to confirm long-term availability and any PCN or EOL plans for the exact suffix before you commit the build. Send us your cold-soak target and annual quantity, and we will quote the E1SB family, including samples and the sibling grades, before you freeze the clock tree.